Jensen Huang calls agentic AI accretive

Software is the domain here. Pricing power is the stake. The gap is that investors have spent the past year treating AI agents as a reason SaaS margins mi(cnbc.com), alongside ServiceNow CEO Bill McDermott, the Nvidia CEO said agentic AI is “fully accretive” for software companies. (cnbc.com)tive” mean here? Basically, Huang is saying agents can add revenue instead of just cannibalizing existing subscriptions. If old SaaS sold access to a (newsroom.servicenow.com)e broadly since at least March, when he said all software will become agentic and companies will “rent out agents to do work.” (constellationr.com) ### Why is that a big shift? Because the bear case on enterprise software has been simple — if AI does the work, customers may need fewer human users, which means few(cnbc.com)ck and charge for specialized digital labor. In his framing, agentic AI is not just a feature upgrade. It is a new business model. (constellationr.com) ### Why does Nvidia care so much? Nvidia wins when companies run more inference, more workflows, and more always-on systems. An agent that actually executes tasks burns more compute than a chatbot that just answers questions. So Huang’s so(constellationr.com)more infrastructure behind the scenes. Nvidia’s GTC 2026 keynote page makes clear that “agentic systems” now sit near the center of its full-stack pitch. (nvidia.com) ### Why was ServiceNow on stage with him? Because ServiceNow is one of the clearest enterprise examples of this thesis. On May 5, the two companies expanded their partnership at Knowledge 2026, pitching governed au(constellationr.com) Project Arc, an autonomous desktop agent, and said its AI Control Tower would be included in Nvidia’s Enterprise AI Factory validated design. (newsroom.servicenow.com) ### Why does governance keep coming up? The catch with agents is that they do things, not just say things. That makes oversight much more important. A chatbot that hallucinates (nvidia.com)ance, HR, or security workflow can create a real mess. So the enterprise sell is becoming two-part — autonomy plus control. ServiceNow is leaning hard into that with AI Control Tower, and Nvidia is happy to wrap that governance layer around the compute layer. (newsroom.servicenow.com) ### Is this really about(newsroom.servicenow.com)e application layer, they keep relevance and maybe expand revenue. If the value collapses into foundation models alone, application vendors get squeezed. Huang is clearly arguing for the first outcome. (constellationr.com) ### What’s the real test from here? Whether customers actually buy software this way. Saying “charge for outcomes” is easy. Building reliable agents that finish work, stay inside policy, and save enough time to justify a new bill is the hard part. But that is the bet now. Not that software disappears — that software turns into managed digital labor. (newsroom.servicenow.com) The bottom line is that Huang is trying to reframe the whole SaaS panic. Agentic AI, in his version, is not the thing that kills software companies. It is the thing that gives them a second revenue engine — if they can ship agents people trust. (cnbc.com)